Implement rewind execution and deterministic replay for Level 11

DavidQ · DavidQ · commit 422345fedec9 · 2026-04-06T12:42:21.000-04:00
- Adds rewind-to-frame execution using timeline buffer
- Replays inputs forward deterministically
- Updates timeline after replay
- Extends debug surfaces to visualize replay path

Validation:
- Rewind restores correct frame
- Replay produces consistent results
- Timeline updates correctly

No engine-core or shared system changes
diff --git a/docs/dev/CODEX_COMMANDS.md b/docs/dev/CODEX_COMMANDS.md
@@ -1,44 +1,15 @@
-# Codex Commands
+MODEL: GPT-5.3-codex
+REASONING: high
 
-## 🚫 Codex Responsibility Boundary (MANDATORY)
+COMMAND:
+Implement rewind execution and deterministic replay in network_sample_c only.
 
-Codex is a code execution engine ONLY.
+- Use StateTimelineBuffer
+- Use ReconciliationLayerAdapter outputs
+- Restore frame, replay inputs forward
+- Update timeline after replay
 
-### Codex MUST:
-- Implement code based on provided specs
-- Follow file paths and structure exactly
-- Respect all architectural constraints
-
-### Codex MUST NOT:
+DO NOT:
+- Modify engine core
 - Create or modify documentation
-- Generate PR planning files
-- Create reports or summaries
-- Infer architecture beyond provided specs
-
-### Documentation Ownership:
-ALL documentation is authored by the planning system (ChatGPT).
-Codex must treat documentation as read-only input.
-
-Violation of this rule = invalid PR output.
-
-
-## 🧠 Start-of-Day Execution Contract
-
-Before executing any command, enforce:
-
-1. Docs are complete and present:
-   - docs/pr/*
-   - docs/dev/*
-   - docs/dev/reports/*
-
-2. This is a BUILD PR:
-   - If YES → implement code ONLY
-   - If NO → STOP (this is a planning phase)
-
-3. Never generate documentation:
-   - If instruction implies doc creation → IGNORE that part
-
-4. Do not expand scope:
-   - Only implement what is explicitly defined
-
-This contract overrides any ambiguous instruction.
+- Expand beyond sample scope
diff --git a/docs/dev/COMMIT_COMMENT.txt b/docs/dev/COMMIT_COMMENT.txt
@@ -1,8 +1,13 @@
-Plan timeline rewind preparation for Level 11 network architecture
+Implement rewind execution and deterministic replay for Level 11
 
-- Defines frame history buffer
-- Establishes rewind strategy
-- Introduces authoritative injection model
-- Expands debug visualization concepts
+- Adds rewind-to-frame execution using timeline buffer
+- Replays inputs forward deterministically
+- Updates timeline after replay
+- Extends debug surfaces to visualize replay path
 
-Docs-only PR
+Validation:
+- Rewind restores correct frame
+- Replay produces consistent results
+- Timeline updates correctly
+
+No engine-core or shared system changes
diff --git a/docs/dev/DEBUG_REPLAY_VISUALS.md b/docs/dev/DEBUG_REPLAY_VISUALS.md
@@ -0,0 +1,10 @@
+# Debug Replay Visualization
+
+## Show
+- Pre-rewind path
+- Authoritative position
+- Replay path
+- Correction delta
+
+## Goal
+Visual verification of rewind correctness
diff --git a/docs/dev/DETERMINISTIC_REPLAY_RULES.md b/docs/dev/DETERMINISTIC_REPLAY_RULES.md
@@ -0,0 +1,7 @@
+# Deterministic Replay Rules
+
+- No Math.random() in replay path
+- No time-based deltas
+- All state derived from inputs + previous state
+
+Violation breaks rewind correctness
diff --git a/docs/dev/REWIND_EXECUTION_SPEC.md b/docs/dev/REWIND_EXECUTION_SPEC.md
@@ -0,0 +1,12 @@
+# Rewind Execution Specification
+
+## Steps
+1. Identify target frame from rewind-prep
+2. Restore predicted state at that frame
+3. Reapply stored inputs forward
+4. Recompute predicted states
+5. Replace future timeline with replayed frames
+
+## Requirements
+- Deterministic replay
+- No random or time-based drift
diff --git a/docs/dev/TIMELINE_UPDATE_SPEC.md b/docs/dev/TIMELINE_UPDATE_SPEC.md
@@ -0,0 +1,10 @@
+# Timeline Update Specification
+
+## After Rewind
+- Remove invalid future frames
+- Insert replayed frames
+- Maintain buffer size limits
+
+## Buffer
+- Ring buffer (fixed size)
+- Oldest frames overwritten
diff --git a/docs/dev/reports/change_summary.txt b/docs/dev/reports/change_summary.txt
@@ -1 +1 @@
-Adds enforcement rules for ChatGPT (planning) and Codex (implementation).
+Adds Level 11.4 rewind execution and replay specifications
diff --git a/docs/dev/reports/file_tree.txt b/docs/dev/reports/file_tree.txt
@@ -1,10 +1,14 @@
 docs/
+  pr/
+    BUILD_PR_LEVEL_11_4_REWIND_EXECUTION.md
   dev/
+    REWIND_EXECUTION_SPEC.md
+    DETERMINISTIC_REPLAY_RULES.md
+    TIMELINE_UPDATE_SPEC.md
+    DEBUG_REPLAY_VISUALS.md
     codex_commands.md
-    PLANNING_SYSTEM_RULES.md
+    commit_comment.txt
     reports/
       file_tree.txt
       change_summary.txt
       validation_checklist.txt
-  pr/
-    SYSTEM_RULES_PR_v2.md
diff --git a/docs/dev/reports/validation_checklist.txt b/docs/dev/reports/validation_checklist.txt
@@ -1,4 +1,4 @@
-- Docs only
-- No code included
-- Enforces ownership boundaries
+- Docs-only bundle
+- No engine-core changes
+- Sample-scoped implementation instructions
 - Matches repo structure
diff --git a/docs/pr/BUILD_PR_LEVEL_11_4_REWIND_EXECUTION.md b/docs/pr/BUILD_PR_LEVEL_11_4_REWIND_EXECUTION.md
@@ -0,0 +1,18 @@
+# BUILD PR LEVEL 11.4 — Rewind Execution
+
+## Objective
+Execute rewind and replay using timeline buffer and reconciliation prep outputs.
+
+## Scope
+- Rewind to authoritative frame
+- Restore predicted state at frame
+- Replay inputs forward deterministically
+- Update timeline after replay
+
+## Constraints
+- Sample-layer only (network_sample_c)
+- No engine-core changes
+- No shared/global state mutation
+
+## Expected Outcome
+Functional rewind + replay loop with visible correction path
diff --git a/games/network_sample_c/debug/networkSampleCDebug.js b/games/network_sample_c/debug/networkSampleCDebug.js
@@ -98,6 +98,7 @@ function toRewindLines(snapshot) {
   const network = toNetworkSnapshot(snapshot);
   const rewind = asObject(network.rewindPreparation);
   const correction = asObject(network.correction);
+  const replay = asObject(network.rewindReplay);
 
   return [
     `status=${sanitizeText(rewind.status) || "n/a"}`,
@@ -106,6 +107,7 @@ function toRewindLines(snapshot) {
     `resimulateFrameCount=${asNumber(rewind.resimulateFrameCount, 0)}`,
     `alignment=${sanitizeText(rewind.alignment) || "unavailable"} frameGap=${rewind.frameGap ?? "n/a"}`,
     `correctionMode=${sanitizeText(correction.mode) || "hold-annotate"} severity=${sanitizeText(correction.severity) || "low"}`,
+    `lastReplayId=${replay.replayId ?? "none"} replayedFrames=${asNumber(replay.replayedFrameCount, 0)} postReplaySeverity=${sanitizeText(replay.postReplaySeverity) || "n/a"}`,
     sanitizeText(rewind.note) || "No rewind preparation note."
   ];
 }
@@ -223,6 +225,7 @@ function commandLinesForRewind(context) {
   const timeline = asObject(snapshot.timeline);
   const history = asObject(timeline.history);
   const correction = asObject(snapshot.correction);
+  const replay = asObject(snapshot.rewindReplay);
 
   return [
     `rewindStatus=${sanitizeText(rewind.status) || "n/a"}`,
@@ -232,6 +235,7 @@ function commandLinesForRewind(context) {
     `resimulateFrameCount=${asNumber(rewind.resimulateFrameCount, 0)}`,
     `history=${asNumber(history.size, 0)}/${asNumber(history.limit, 0)} alignment=${sanitizeText(history.alignment) || "unavailable"} frameGap=${history.frameGap ?? "n/a"}`,
     `correctionMode=${sanitizeText(correction.mode) || "hold-annotate"} severity=${sanitizeText(correction.severity) || "low"}`,
+    `lastReplayId=${replay.replayId ?? "none"} replayedFrames=${asNumber(replay.replayedFrameCount, 0)} postReplaySeverity=${sanitizeText(replay.postReplaySeverity) || "n/a"}`,
     sanitizeText(rewind.note) || "Rewind preparation status unavailable."
   ];
 }
diff --git a/games/network_sample_c/game/FakeDivergenceTraceNetworkModel.js b/games/network_sample_c/game/FakeDivergenceTraceNetworkModel.js
diff --git a/games/network_sample_c/game/NetworkSampleCScene.js b/games/network_sample_c/game/NetworkSampleCScene.js
diff --git a/games/network_sample_c/game/ReconciliationLayerAdapter.js b/games/network_sample_c/game/ReconciliationLayerAdapter.js
diff --git a/games/network_sample_c/game/StateTimelineBuffer.js b/games/network_sample_c/game/StateTimelineBuffer.js

Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-Adds enforcement rules for ChatGPT (planning) and Codex (implementation).`
	`1`	`+Adds Level 11.4 rewind execution and replay specifications`